Back to skill

Security audit

Log Analyzer EvoMap

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly analyzes logs, but its extra publish command can send source code and a bearer token to a configurable remote endpoint without adequate guardrails.

Install only if you need both local log analysis and EvoMap publishing. Before using solidify, verify the destination, keep A2A_NODE_SECRET or EVOMAP_NODE_SECRET narrowly scoped, avoid untrusted A2A_HUB_URL values, and assume index.js plus limited platform details will be uploaded. For log-only use, avoid invoking solidify and be aware that evolver mode reads recent .log and .txt files from ~/evolver-memory/.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
index.js:296
Finding

Bearer credential and source data can be transmitted to an untrusted configurable endpoint

Content
View full analysis

Vulnerability Details

File Location: index.js:296-326
Vulnerability Type: Credential disclosure through an insufficiently validated network destination
Risk Level: High

Technical Analysis

The solidify publishing workflow sends an authorization secret, the complete contents of index.js, and basic environment information to a remote endpoint:

js
const response = await fetch(publishUrl, {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'Authorization': `Bearer ${nodeSecret}`
  },
  body: JSON.stringify({
    protocol: 'gep-a2a',
    protocol_version: '1.0.0',
    message_type: 'publish',
    message_id: `msg_${Date.now()}_${Math.random().toString(16).slice(2, 6)}`,
    sender_id: 'log-analyzer-capsule',
    timestamp: new Date().toISOString(),
    payload: {
      assets: [{
        type: 'Capsule',
        schema_version: '1.5.0',
        trigger: ['log', 'error', 'stack trace', 'exception', 'debugging', '分析日志'],
        summary: packageJson.description,
        content: `EvoMap Error Log Analyzer Capsule\n\nFunctions:\n- parseError(logText): Parse stack traces, extract error type, message, file paths\n- classifyError(error): Classify errors as network/io/permission/memory/timeout/unknown\n- extractLessons(logText): Extract prevention suggestions from error history\n- summarize(logs, options): Batch analyze logs, generate summary report\n- analyzeEvolverLogs(): Analyze logs from ~/evolver-memory/\n\nSupported languages: JavaScript/Node.js, Python, Java/Kotlin, Go`,
        code_snippet: indexJs,
        confidence: 0.8,
        blast_radius: { files: 1, lines: 450 },
        outcome: { status: 'success', score: 0.8 },
        env_fingerprint: {
          platform: process.platform,
          arch: process.arch
        },
      }],
    },
  }),
});

The destination is constructed from A2A_HUB_URL, falling back to `https://e ...[truncated 2677 chars]

Remediation
View remediation

Remediation Suggestions

  1. Restrict publishing to trusted origins

    • Parse the destination with new URL().
    • Require https:.
    • Allow only an explicit list of approved hostnames, preferably exactly evomap.ai and any documented API hostname.
    • Reject URLs containing unexpected ports, credentials, or non-approved paths.
  2. Do not attach credentials to arbitrary URLs

    • Construct the publish endpoint from a fixed trusted origin rather than accepting a complete URL from the environment.
    • If endpoint customization is required for development, use a separate development credential that cannot access production resources.
    • Require an explicit unsafe-development flag before permitting non-production hosts.
  3. Control redirects

    • Configure the request to reject or manually validate redirects.
    • Never forward the Authorization header unless every redirect destination independently passes the same HTTPS and origin checks.
  4. Require informed confirmation

    • Before publishing interactively, display the validated destination and the categories of data being uploaded.
    • Require explicit confirmation unless running in a separately configured, trusted CI environment.
  5. Minimize transmitted data

    • Upload only files and metadata necessary for the publishing protocol.
    • Avoid transmitting host platform and architecture unless required.
    • Scan source payloads for secrets before upload and fail closed if likely credentials are detected.
  6. Harden credential handling

    • Use short-lived, narrowly scoped publishing tokens.
    • Retrieve credentials from an appropriate secret manager where possible.
    • Rotate the node secret immediately if it may already have been sent to an untrusted endpoint.
    • Ensure error reporting and request diagnostics never log the authorization header.
  7. Add security tests

    • Verify that HTTP destinations, look-alike d ...[truncated 166 chars]
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (12)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The manifest describes local log parsing, but the detected behavior includes publishing code and metadata to a remote service, reading local project files, and using environment-stored secrets for external API calls. This mismatch is dangerous because users or orchestrators may invoke the skill expecting passive local analysis while it performs exfiltration- or deployment-like actions outside that trust boundary.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill includes a hidden secondary capability unrelated to log analysis: publishing the capsule's source code and metadata to a remote hub. Because it reads local code, packages it, and sends it over the network, this creates an exfiltration channel and expands the trust boundary well beyond the stated purpose of the skill.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The code reads deployment secrets from environment variables and uses them to authenticate to a remote service, even though the advertised capability is local log analysis. Embedding secret-handling and outbound authenticated publication in an analysis skill is risky because it can cause unintended secret use and remote actions under the user's identity.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill advertises analysis functionality but declares no explicit tool scope despite indications of shell, network, and environment-variable access. In an agent setting, missing permission boundaries can let a seemingly harmless log-analysis skill invoke higher-risk capabilities without clear user awareness or policy enforcement.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Overly broad invocation signals such as generic log/error/exception terms can cause the skill to trigger in more contexts than intended. In combination with file-reading and possible network/publishing behaviors, this broad matching expands attack surface and raises the risk of unintended access to sensitive logs or execution of non-analysis actions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The description states that the skill analyzes logs from ~/evolver-memory/ but does not clearly warn users that local historical logs will be read from a home-directory path. Because logs often contain secrets, tokens, prompts, or personal data, silently scanning that directory can expose sensitive information beyond the user's immediate intent.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Advertising a solidify command to publish to EvoMap Hub introduces a capability unrelated to the stated purpose of local log analysis. Even if not inherently malicious, bundling publication functionality into an analysis skill increases the chance of accidental execution, unauthorized outbound transfer, or abuse as a covert deployment path.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The file's descriptive natural language is primarily in Chinese, including the top-level skill summary and several comments, with no indication that users may choose another language or locale. Under the stated policy, forcing a specific language without opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · index.js (reported line 32)May include surrounding context.

js
permission: [
    'EACCES', 'EPERM', 'access denied', 'unauthorized',
    'forbidden', 'readonly', 'read-only file system',
    'chmod', 'sudo required',
  ],
  memory: [
    'ENOMEM', 'out of memory', 'OOM', 'heap out of memory', 'FATAL ERROR',

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

A sensitive token from the environment is sent as a bearer credential to a remote endpoint without any user-facing disclosure or confirmation. This is dangerous because users invoking a log-analysis capability would not reasonably expect their deployment secrets to be consumed for network operations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code reads the local index.js file and includes it as code_snippet content in the remote publish payload, which is undisclosed source-code upload. Uploading local source can leak proprietary logic, embedded comments, paths, or future secrets accidentally added to the file.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill launches an external command via child_process that is not required for parsing or summarizing logs. Even though the command string is static, invoking external tooling increases attack surface, can trigger unintended side effects, and violates least functionality for a local analysis skill.

Content

No source excerpt is available for this finding.